keras-team/keras · error · ValueError

`input_tensors` must have the same structure as…

Error message

`input_tensors` must have the same structure as model.input\nReference structure: {model.input}\nReceived structure: {input_tensors}

What it means

Structural guard in _clone_functional_model: the provided input_tensors must mirror the exact nesting structure of model.input (same dict keys / list / tuple shape). After flattening, layers are zipped positionally with the placeholders, so a mismatched structure (e.g. a flat list for a dict of inputs, or wrong nesting depth) would silently wire inputs to the wrong layers and is rejected instead.

Solutions

1. Mirror model.input's structure exactly: pass a dict with the same keys if the model was built from dict inputs, or a list with the same length and nesting
2. Use tree.structure(model.input) as a reference, or omit input_tensors to let placeholders be generated automatically

When it happens

Trigger: Thrown at keras/src/models/cloning.py:392 when the library encounters an invalid state.

Common situations: See trigger scenarios.

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/87610939f19e7c45. Report an issue: GitHub.

Appendix: source

Thrown at keras/src/models/cloning.py:392

    if not isinstance(model, Functional):
        raise ValueError(
            "Expected `model` argument "
            f"to be a Functional Model instance. Received: model={model}"
        )

    if input_tensors is not None:
        if not all(
            isinstance(x, backend.KerasTensor)
            for x in tree.flatten(input_tensors)
        ):
            raise ValueError(
                "All entries in `input_tensors` must be KerasTensors. "
                f"Received invalid values: inputs_tensors={input_tensors}"
            )
        try:
            tree.assert_same_structure(input_tensors, model.input)
        except ValueError as e:
            raise ValueError(
                "`input_tensors` must have the same structure as model.input"
                f"\nReference structure: {model.input}"
                f"\nReceived structure: {input_tensors}"
            ) from e
    else:
        input_tensors = tree.map_structure(
            lambda x: Input(batch_shape=x.shape, dtype=x.dtype, name=x.name),
            model.input,
        )

    def operation_fn(layer):
        new_layer = clone_function(layer)
        return new_layer

    output_tensors = model._run_through_graph(
        input_tensors,
        operation_fn=operation_fn,
        call_fn=call_function,

View on GitHub (pinned to 7a34a03db6)